ESMC: Entire Space Multi-Task Model for Post-Click Conversion Rate via Parameter Constraint

• DOI: 10.48550/arxiv.2307.09193
• Publication Date: 2023-01-01
• Authors: Zhenhao Jiang; Biao Zeng; Hao Feng; Jin Liu; Jicong Fan; Jie Zhang; Jia Jia; Ning Hu; Xingyu Chen; Xuguang Lan
• Keywords: Click-through rate; Tracing

Large-scale online recommender system spreads all over the Internet being in charge of two basic tasks: Click-Through Rate (CTR) and Post-Click Conversion (CVR) estimations. However, traditional CVR estimators suffer from well-known Sample Selection Bias Data Sparsity issues. Entire space models were proposed to address issues via tracing decision-making path "exposure_click_purchase". Further, some researchers observed that there are purchase-related behaviors between click purchase, which can better draw user's intention improve recommendation performance. Thus, has been extended "exposure_click_in-shop action_purchase" be modeled with conditional probability approach. Nevertheless, we observe chain rule does not always hold. We report Probability Space Confusion (PSC) issue give a derivation difference ground-truth estimation mathematically. propose novel Multi-Task Model for Parameter Constraint (ESMC) alternatives: Siamese Network (ESMS) Global Domain (ESMG) PSC issue. Specifically, handle action" "in-shop separately light characteristics in-shop action. The first is still treated while second one parameter constraint strategy. Experiments on both offline environments large-scale illustrate superiority our methods state-of-the-art models. real-world datasets will released.